Building Strongly-Reliable and High-Performance Storage Cluster with Network Attached Storage Disks Using PCs

Abstract

Decentralized, cooperative and large-scale distributed storage systems that consist of a cluster of storage nodes attached with local disks can deliver high resource utilization, high availability and easy scalability. This paper describes the design and prototype implementation of a strongly reliable and high-performance storage cluster using commodity components, where each node is equipped with local storage disks. We present a self-adaptive redundancy scheme for file data access that uses a combination of RAID5 and RAID1 writes to store data. To solve the disks latency issue, we adopt a novel cluster caching scheme that shares memories between nodes in cluster to construct an efficient and cooperative cache-to-disk accesses policy. Performance measurements of such a system are encouraging -showing that such a storage cluster dynamically provide the high throughput and the strong reliability.